DBG2OLC: Efficient Assembly of Large Genomes Using the Compressed Overlap Graph

The highly anticipated transition to the third generation DNA sequencing (3rdGS) technology have reached a stalemate primarily due to the high error rates (15-45%), which make the assembly of long erroneous reads extremely challenging because existing software solutions for 3rdGS assembly are often overwhelmed by error correction tasks. We report three significant breakthroughs that push the envelope of genome assembly and offer an enabling software solution to overcome the current 3rdGS stalemate. Firstly, we take a counter-intuitive strategy and develop a base-level correction-free assembly algorithm, which resorts to data compression technology and the assembly was performed with the compressed reads. Magnitudes of compression lead to magnitudes of reduction in read lengths, enabling magnitudes of savings in computational time and memory space. We implement the new algorithm in a proof-of-concept software package DBG2OLC. Experiments with the 3rdGS data including PacBio and Oxford Nanopore show that our method is able to assemble large genomes magnitudes more efficiently than existing methods. For example, on a large PacBio human genome dataset we calculated the all-pair alignment of 54x erroneous long reads in 6 hours compared to the 405,000 CPU hours previously reported by Pacific Biosciences. Secondly, while maintaining comparable high quality assemblies, our approach requires significantly lower sequencing coverage (10x-20x) than existing assemblers, which translates to significant cost-cut for genome sequencing. Thirdly, our method is highly adaptive and is the only one to date that demonstrates ultra efficiencies not only for the 3rdGS PacBio and Nanapore sequences, but also for the latest NGS data.